Image Scaler Wan 2.2 I2V MXD

Wan 2.2 I2V expects its start image to land in one of a handful of resolution buckets, and getting there by hand means juggling aspect ratios, multiples-of-whatever, and "does 1280x720 actually count?" every single time. This node is the one-drop answer: feed it your start image, pick a tier, and it returns the correctly bucketed version - no padding, no math homework.

It's the Wan 2.2 sibling of the pack's LTX scaler, tuned for the two-pass I2V workflow. The "No Pad" in the source's own title is the philosophy: rather than letterboxing an image into a bucket, it resizes to fit or crops to cover, so the model gets clean pixels instead of a band of black.

How it works

tier picks your target resolution class: Auto, 480p, or 720p (the default is Auto, which maps to the legacy "Safe Auto" behavior - if your input is already a valid Wan 2.2 bucket it passes straight through; if it's far outside the 480p–720p range it errors early instead of producing garbage; otherwise it behaves like Auto). Square-ish inputs get routed to the square buckets - 624×624 for Auto/480p, 1024×1024 for 720p.

crop_to_fit is your mode switch:

• on (default) - resize-to-cover then center-crop to the exact target bucket.
• off - proportional resize, no crop, may come out smaller than the target. Better if you refuse to lose pixels, but the latent will be smaller.

aspect_mode lets you force the bucket shape - Auto picks wide/tall/square from the source, or you can pin Tall, Wide, or Square when you want a specific composition regardless of input.

Inputs and output

• image - your I2V start image.
• tier - Auto / 480p / 720p.
• crop_to_fit - boolean.
• aspect_mode - Auto / Tall / Wide / Square.

Output: the scaled IMAGE, ready to feed the Wan 2.2 image-to-video encode.

Installing it

Part of the ComfyUI-MaxedOut pack. ComfyUI Manager search Maxed Out, or:

cd ComfyUI\custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

Restart ComfyUI. No extra dependencies, no model downloads.

Where people get burned

The crop-vs-fit decision is the one that actually changes your output. Crop gives exact buckets (best for quality-focused I2V) but cuts your edges; fit keeps everything but leaves you under-bucket. And don't fight the square routing - feeding a square image with aspect_mode forced to Wide will crop aggressively, so let Auto handle it unless you specifically want that composition. Finally, remember this node only scales the image; the bucket-aligned latent still comes from your Wan 2.2 empty-latent setup, so keep the two consistent.